Claude + Lean formally verifies Agent SDK, yielding 16 bug-fix PRs from short prompts
JiaweiLiu_ · x · 2026-09-24
Anthropic engineer Boris Cherny shares that he used Opus 5.5 with Lean to formally verify the Claude Agent SDK — a couple of short prompts produced 16 PRs fixing bugs and race conditions. He sometimes combines Lean and TLA+ to hunt issues in data flow, concurrency, and state management, noting Claude excels at both despite his own unfamiliarity.
Related ecosystem work from the TLA+ community and Specula team:
- Specula: an agentic tool that models system code in TLA+ and uses model checkers to surface hundreds of deep bugs (480 stars on GitHub)
- TLAPS-Bench: a proof service for TLA+ specs; AI proved classic protocols like 2PC, Paxos, and TCP, forcing the team to retire all proof-completion tasks as too easy
- Proving low-level specs of real-world system code from scratch remains non-trivial
More from coding & agent
- User automates entire video clipping workflow with Nous Research's Hermes agent — Teknium · 2026-11-12
- As models improve, writing good evals gets harder and more important — amankhan · 2026-11-12
- Autonomous agent edits just 1,024 weights to lift Qwen3.5 0.8B from 7.4% to 37.4% on MATH-500 — VictorSanh · 2026-11-12
- Free AI game jam in SF Nov 14–15 with teams from Anthropic, Google, Meshy, OpenAI — chrisfirst · 2026-11-12
- Omarchy 4.6 lets you share local AI models across your tailnet, says DHH — alexcovo_eth · 2026-11-12
- Coding harnesses hand shell access to inference providers — 4-step gateway mitigation — shensi · 2026-11-12